Incident detection using the Standard Normal Deviate model and travel time information from probe vehicles

Abstract

One application of travel time information explored in this thesis is freeway incident detection. It is vital to develop reliable methods for automatically detecting incidents to facilitate the quick response and removal of incidents before they cause breakdowns in traffic flow. The use of real-time travel time data to monitor freeway conditions has the advantages over conventional loop detectors of taking into account the dynamic, longitudinal nature of traffic flow and requiring data from only a portion of the traffic stream. This study employed the Standard Normal Deviate (SND) Model to test the feasibility of using travel time data to detect lane blocking incidents. The fundamental concept of the SND Model was based on the comparison of real-time travel time data to historical travel time data for given freeway segments during specified times. The travel time and incident reports used were collected through the Real-Time Traffic Information System (RTTIS) in the north freeway corridor of Houston, Texas using probe vehicles equipped with cellular telephones. The data were compiled on 39 freeway links from October 1991 through August 1992 on weekdays during morning and afternoon data collection periods. The results of incident detection tests, applying the SND Model to incident and travel time me data from the North Freeway, indicated high successful incident detection rates. However, high false alarm rates also resulted from the SND Model test applications. An optimum SND value of 2.0 was observed for the North Freeway test data. At this value the SND tests produced successful incident detection rates of 70 percent and higher during both the morning and afternoon periods. False alarm rates were also 70 percent. The best results were achieved on those freeway sections where the most incident and travel time data had been collected. The overall results of the incident detection tests on the North Freeway demonstrated that the SND Model was a feasible incident detection algorithm, but required an extensive historical travel time data base.